Prim Bonuses and LODs

[Aquila Kytori]

I hadn't thought of only using 3 triangles to represent each stack and felt sure using the version where the tris are joined at the pointy end would make a noticeable difference compared to the Lowest Lod that I had been using.

Unfortunately not

Note: Triangle and vertex count are those given in the Uploader and both LOD0a and LOD0d are smooth shaded.



 

If I really needed to reduce the download cost more then I would have to experiment with using less geometry in the other  Lod meshes.

Using the "2 triangle" example (like the "2 plane" version in the image above, LOD0b) would really mean using 4 because of the need to duplicate and flip the normals so that the texture could be seen from all angles.

[ChinRey]

---

Aquila Kytori wrote:

I hadn't thought of only using 3 triangles to represent each stack and felt sure using the version where the tris are joined at the pointy end would make a noticeable difference compared to the Lowest Lod that I had been using.

---

Your lowest LoD model is so simple to start with, there is not much to gain anyway but it's a very good example to illustrate some of the tricks of advanced LoD optimisation.

The one that should give the lowest download weight is the one where you simply triangulate the quads and remove the lower tri from each pair. I included the other examples since there was a chance they might be more visually satisfying without causing a significant LI increase.

---

Aquila Kytori wrote:

Using the "2 triangle" example (like the "2 plane" version in the image above, LOD0b) would really mean using 4 because of the need to duplicate and flip the normals so that the texture could be seen from all angles.

---

Yes, that too was included because of its possible visual advantages. You should be able to get the download weight below 1.5 if you close the gaps between the sheets and make sure they are aligned precisely along the x and y axises.

 

Edit: Maybe I should illustrate what I mean with aligning precisely:



A challenge to everybody: what is the difference between these two models that causes one of them to have 50% higher download weight than the other?

[Aquila Kytori]

 

uuuuuuuuuuumm................ would the clue to the answer be found in the quote below ? :

---

ChinRey wrote:

Edit: Maybe I should illustrate what I mean with

ALIGNING

precisely:

---

 Just a guess, could it be that maybe, perhaps or be that possibly the first has faces aligned exactly to X or Y and Z axis ?

 

[IvanBenjammin]

I've done some testing: I made a cylinder in Maya, 1 version with split edges, the other bevelled. In both cases, they have Maya's soft normals (which aren't quite the same thing as averaged normals) the UVs were reduced down to a single point. I import to Blender as obj, then output as dae.

In both cases, I got the results I was expecting. With split edges, the SL uploader was telling me it had 128 triangles, 130 vertices (the extra 2 vertices accounted for because of the ends of the cylinder). The bevelled version (with both top and bottom edge loops bevelled) had 256 triangles, 130 vertices.

Now, it gets interesting when I start delving into Blender's smoothing functions. I have to export from Maya without normal data to get Blender's auto smoothing to work, but when I exported the bevelled cylinder with auto smooth at 180 degrees, SL's uploader tells me it has 256 triangles, 258 vertices. Without auto smoothing (and without imported normals from Maya) its back to 256 triangles, 130 vertices.

So it would seem from my testing that auto smoothing is indeed the culprit.

[ChinRey]

---

Aquila Kytori wrote:

Just a guess, could it be that maybe, perhaps or be that possibly the first has faces aligned exactly to X or Y and Z axis ?

---

You figured it out right away???

Now let's see if anybody else can guess. And no cheating! Don't look at Aquila's answer!

---

Seriously, this is a rather extreme example but a 20-30% change in download weight depending on the alignment of the vertices in the model and on the UV map is not unusual at all.

Unfortunately, One of the invetibable consequences of that is going to be very, very unpopular:

• Never ever use a modifier in Blender to alter the geometry of your model!

Not if you want the best possible LI that is. The reason si that it'll take so long to clean up the file after the modifier you're always better off doing the job manually.

I actually never work in Object Mode in Blender at all. I go to Edit Mode and give each and every vertice the personal attention they deserve.

Edit: forgot to actually answer Aquila's question:

---

Aquila Kytori wrote:

Just a guess, could it be that maybe, perhaps or be that possibly the first has faces aligned exactly to X or Y and Z axis ?

---

Well, an equal sided hexagon can't really have faces aligned to both x and y axis at the same time of course and both models are precisely aligned on z (if I had changed that too, download weight might well have ended up at 1.2) but yes, you got the idea.

[arton Rotaru]

Mine comes in at 0.255. :matte-motes-tongue:



[ChinRey]

---

arton Rotaru wrote:

Mine comes in at 0.255. :matte-motes-tongue:

---

Interesting. I can't see a way to achieve that number...



I mean exactly that number  of course

 

On the other end of the scale:

 



 

I couldn't get it all the way up to 1.2. Sorry.

 

And a final one - this is what's technically known as cheating (or "no UV map" if you like):



Come to think fo it, maybe I can see how Arton got a slightly different base value than me, different UV mapping could do that

[arton Rotaru]

0.184 :matte-motes-delicious:





[ChinRey]

Very interesting. I didn't keep the file so I can't show my UV map but it wasn't actually that different. I had the ends joined to the last quad in the row so the row didn't extend all the way across the grid. I suppose that's the explanation.

I would think the most LI efficient UV map would be to have the side and the ends mapped separately, each covering the whole grid but I'm not going to test it because this has given me another important point to make.

Arton's and my examples show one way a skilled SL mesh maker can manipulate the LI without resorting to LI butchery but they also illustrate another perhaps even more important point: All these examples are in full LoD, that is all LoD models are identical. That means no breakdown at any distance and no Li changes when the object is resized. And even so, no matter how hard I tried  to maximize the LI but I simply couldn't get it up to 1.

I've seen people who regard themselves as skilled mesh makers upload simple cubes with poor LoD. Now, you can of course ask what (if anything) goes through the mind of somebody who believes a mesh cube in SL is a good idea at all but even so, how on earth do they manage to mess up the LoD??? This is the level of ignorance we see far too often. And, just to avoid any misunderstanding, I'm not talking about the many unpretentious people who find joy putting things together in SL, I'm talking abut people who call themselves experts, market their builds as "high quality" and expect to be paid quite substantial amounts of Lindens for it even though they fail to graps even the most basic principles of the craft they pretend to be masters of.

I couldn't mention names here even if I wanted to (which I don't) but there is a new house on my land now. The tenant wanted it and it's hidden way up on a sky platform so OK, I don't mind. It's a dodgy build in any way but the most noticeable detail is the house control panel. It's a flattened mesh cube. That is, it's a mesh cube if you cam in on it, at any reasonable viewing distance it's a mesh triangle. I got a bit curious so I checked the maker's profile and it was full of EULA, what you were allowed to do and especially what you were not allowed to do with his masterworks (which incidentally aren't even modifialbe and certainly not transferable). Checked out MP, that house costs 1495 Lindens, it's marketed as low LI even though it weighs in at a whopping 18 (for a very simple structure with horrible LoD) and it has eight 5 star reviews, including some rather recent! The maker obviously can't make a proper mesh to save his life yet he's been able to fool enough of the people he had foisted his garbage onto to believe it was top notch!

This is not an unusual story and it's what ruins Second Life and causes most of the mesh disasters. It's not the people who read and contribute to this forum. Everybody here are eager to learn how to improve our skills no matter how good or poor they are at the moment. If we weren't we wouldn't be here. The problem isn't of all the unskilled but unpretentious builders either. The problem are all those "builders" who think they already know everything and are so full of themselves they don't realize the only thing they've mastered is ignorance.

Oh well, this turned into a rant. Sorry about that.

Here's another suggestion for Patch, assuming he read my first one but couldn't find a time machine:

Go to Meauxle Bureaux and delete it. If you really want to promote quality building, you just can't afford to have a textbook example How Not To Make Mesh as the "Home of the Linden Department of Public Works". Besides, some of the people who committed items to that monstrosity have later grown into quite decent mesh makers and they really don't deserve to have their beginners' mistakes displayed that way.

[Chic Aeon]

It is certainily likely and quite possible that those reviews aren't actual reviews at all. It is a long time practice to "seed" reviews with alts and friends. It is also a long time practice to buy competitor's items and give one star scathing reviews ^^. In many ways reviews don't count for anything in SL and that is a sad case.

This is talked about often on the merchant board. I have one "person" who I am pretty sure I know the identity of who continues to buy a product apparently just so that he (she, they, alt 3 etc) can get the rating down to 1.  I know an expensive large home make who was notorious for instant and glorious reviews on new houses LOL which were not all that great.

OH, and the item that keeps getting one star reviews (I think there were three or four this last month - on a rampage) has a FREE DEMO and has had for well over a year. So there is no reason for anyone to buy something they have issues with and give a 1 star review.

 

Oh wait -- there is.

 

Sorry. Off topic but it really makes me mad.   Pertanent comment in bold

I doubt seriously that the person on your land would have bought the house if his LOD wasn't set up so high. We will be living with the aftermath of "how to make your scuplties look better" notecard instructions for a long while I am afraid.

[ChinRey]

---

Chic Aeon wrote:

It is certainily likely and quite possible that those reviews aren't actual reviews at all.

---

It's possible but in this case I doubt it. The reviews were well spread out in time for a start.

A while ago I stumbled across another piece of crap with a suspiciously large amount of five star reviews and decided to investigate a bit. I looked at the profiles of the reviewers and couldn't find anything that suggested foul play. Genuine SL'ers with very different ages and parts of very different inworld groups so I'm fairly sure they wrote those reviews in good faith. However, some of them had their home listed in their picks and I went and had a look. None of them was actually using the house they had given such a positive review of. I guess the love didn't last.

 

---

Chic Aeon wrote:

I doubt seriously that the person on your land would have bought the house if his LOD wasn't set up so high.

---

It's probably more like: see pretty picture on MP ... buy.

The LoD isn't actually a problem in this particular case. It's a very small sky platform with fairly tall screens around it so the basic structure will always show in full LoD. This is the reason why I allow the house on my land. I don't even want to think of how it would work in a normal more open environment though.

 

---

Chic Aeon wrote:

We will be living with the aftermath of "how to make your scuplties look better" notecard instructions for a long while I am afraid.

---

Yes, that's another of the Seven Big Blunders I mentioned earlier.

[Chic Aeon]

I certainly don't understand why folks buy things from a photo IF there is a demo out or even in my one star review case FREE TO TRY.   I have only purchased a very few things without seeing and they were under $100 linden so no big deal if awful.

 

I was coming to post a screenshot that I just took so I will paste it in here. This is one of the works of a popular creator (one of a FEW mind you) that seem to give no thought to uploading. The items are very nice but fall apart before they even rez. And while they ARE 1 land impact (a seemingly magic number) they could have MUCH BETTER HOLDING with no real increase in the LI number. It boggles my mind really.

 



 

I am pretty sure you know who this if from and if not from that person, then from one in the same mindset. It is not my place to preach to them; they are much more fa-mous than I and certainly make more money (or I would suspect so) but still, I cringe. A couple of extra clicks and people not running LOD 4 could SEE this stuff (sigh).

 

Great new notebook-tablet computer today. We have already bonded -- and I made some ART, so a good day.

 

 

[arton Rotaru]

It went down to 0.071 with the UVs laying on the borders of the 0 to 1 UV space. :matte-motes-wink:

 

---

ChinRey wrote:

 how on earth do they manage to mess up the LoD???

---

Good question. :matte-motes-little-laugh: Perhaps one of those who also recommend to set RenderVolumeLODFactor to 4 or higher even.

 

 

[Medhue Simoni]

Well, I did finally make another LOD video. This 1 was just about the uploader and generating LODs. Don't know how happy I am with it tho. Everytime I tried, it just seemed like I was almost promoting the 1 triangle cheat, which bothered me. I guess tho, there is no way around it, if you are going to talk about the uploader, generating LODs, and getting a low Land impact score.

[Chic Aeon]

REALLY love your elephant -- and the eyebrow movement is fantastic LOL.

 

As we have seen on this thread there are many methods that can be utilized to get good LODs and low land impact numbers.

 

That being said, one thing that wasn't shown in this video and I think really needs to be shown (maybe next video) is how the uploader "previews" relate to the models in world and how they look at the distances based on the size of the models (in this case medium furniture size). I don't think anyone not intimately involved already with uploading mesh will "get" the relationship between the viewing pane and the inworld model until it is shown.

 

On that note and taking a segue, I tried on a pair of very cute bootlets this evening. They looked great. I did note however a HUGE jump in my avatar rendering cost as they placed themselves on my tootsies. Took them off, put them on . Went "wow". Then I rezzed them (and happily they would rez rather than giving an error message).

 

They were 128 land impact!!!!!!!!!

 

We haven't really said much about clothing on this thread, but just because one might make wearables, that doesn't mean they can (should) forget about land impact.  I won't be blogging about those shoes even though they were very cute.  Insert sad face.

                                               ****************************************

There has been a ton of good information on this thread. Lots of opinions that readers can pick and choose from depending on their outlook and perhaps customer base. But even if the only real effect is to get some folks to realize they need to think about LODs, that's a good thing in my books.

 

Pssst!  If there isn't a baby elephant yet, there really needs to be one!

 

[Medhue Simoni]

---

Chic Aeon wrote:

REALLY love your elephant -- and the eyebrow movement is fantastic LOL. 

Pssst!  If there isn't a baby elephant yet, there really needs to be one!

 

---

The elephant is using the bento skeleton, which means users can customize the shape to be much smaller. Almost baby size. That said tho, originally the elephant was made for the Unity3D engine. In Unity, you can use shapekeys. Using shapekeys, I made the african elephant morph into an asian elephant, and both of their baby elephants. So, it would not take much to make the baby african elephant. I'd just have to scale that mesh up to fit the big elephant, and then scale the rig down to be baby size, and export him. The babies are pretty cute.



---

Chic Aeon wrote:

 

 

That being said, one thing that wasn't shown in this video and I think really needs to be shown (maybe next video) is how the uploader "previews" relate to the models in world and how they look at the distances based on the size of the models (in this case medium furniture size). I don't think anyone not intimately involved already with uploading mesh will "get" the relationship between the viewing pane and the inworld model until it is shown.

 

---

Hmm, not quite sure what you're talking about here, but I don't really make furniture.

 

---

Chic Aeon wrote:

 

 

On that note and taking a segue,

I tried on a pair of very cute bootlets this evening. They looked great. I did note however a HUGE jump in my avatar rendering cost as they placed themselves on my tootsies. Took them off, put them on . Went "wow". Then I rezzed them (and happily they would rez rather than giving an error message).

 

They were 128 land impact!!!!!!!!!

 

We haven't really said much about clothing on this thread, but just because one might make wearables, that doesn't mean they can (should) forget about land impact. 

                                               ****************************************

 

 

---

Well, clothing is no different than avatars. With Bento coming, if I can find the time, I'll be doing videos about Bento, and efficiency of avatars will be a pretty important part of that, as avatars have an avatar complexity score. My elephant is only around 6000 triangles, with a complexity score of 9500. This is literally only like 3000 more than the default, which is ridiculously low. I just rezzed the elephant on the ground, and he is only 90 land impact. He is HUGE tho.

 

---

Chic Aeon wrote:

I won't be blogging about those shoes even though they were very cute.  Insert sad face.

                                               ****************************************

 

 

---

You have a blog? Can you give a link?

[Arthur Reece]

It's wonderful that Linden Labs is increasing the prim count by 33% to allow creators to import mesh with a higher LOD and therefore improve the look...but...what they fail to mention is that uploading with a higher LOD physics shoots the cost of the uploads through the roof! What costs L$11 to upload with Medium physics jumps to over L$40 at the High physics...great way to increase their income through the back door

[arton Rotaru]

---

Arthur Reece wrote:

It's wonderful that Linden Labs is increasing the prim count by 33% to allow creators to import mesh with a higher LOD and therefore improve the look...but...what they fail to mention is that uploading with a higher LOD physics shoots the cost of the uploads through the roof! What costs L$11 to upload with Medium physics jumps to over L$40 at the High physics...great way to increase their income through the back door

---

Well, creating simplified custom physics shapes should be standard regardless of land impact, and LOD models. Physics shapes are pretty much the easiest part of game asset creation. Super simple geometry, no UV mapping, no texturing, no shading.

But you bring a valid vector to the equation which I never thought about. Upload cost. That could be another reason why some people skip LOD models. Though, if upload cost is really a reason to hamper models from being shown as good as they can, that would be really sad.

[Medhue Simoni]

Here's an interesting side note. Unreal is streaming right now and showing off their LOD generator. At least, I think that is what it is.

[ChinRey]

---

arton Rotaru wrote:

But you bring a valid vector to the equation which I never thought about. Upload cost. That could be another reason why some people skip LOD models. Though, if upload cost is really a reason to hamper models from being shown as good as they can, that would be really sad.

---

There may well be people who try to save upload cost by skimping on the LoD. Compared to some of the other fallacies we've mentioned here, that one seems downright plausible. Download weight doesn't really affect upload cost very much and, as we have shown here and in several other threads on the forum, poorly made LoD models doesn't really save downlaod weight anyway.

Sassy once said that any upload cost higher than 11 L$ was a clear sign of a faulty mesh. I thnk that's a bit extreme but 12-20 L$ upload cost definitely means you should reexamine the mesh before uploading. Higher than 20 means there's bound to be something wrong somewhere, possibly with the exception of large fitted meshes and other unusually complex objects that jsut can't be split up in a sensible way.

 

[Chic Aeon]

Wow that is a lot of comments. Let me see if I can comment on your comments to MY comments.

 

Glad to know the elephant can be a baby. I personally have no real desire to be a none human unless maybe for a film which I don't do as much these days, but SUPER cute and impressive.

 

What I meant was that in this thread (and also on the Kettle Quest video from long ago that I mentioned), folks have shown what happens to the models in world when viewed. I am guessing that newish 3D modlers will have a difficult time relating the previews of what the uploader is planning on doing LOD wise to the DISTANCE that relates to inworld. Maybe others will have some ideas on this -- haven't read down the thread yet.

 

The very cute shoes could easily have been lower -- well TONS lower. I have some very detailed slippers (never could wear them as I didn't have Slink flat feet but did REZ them and had them in photos). They were 1 li each. So maybe the designer just uploaded everything as high as it could go?  Have no clue but 128 for a pair of shoes is too much. I think my ARC went up 20,000 or so putting them on or off. Sculpts often do the same thing and those are getting deleted mostly. A few low impact jewelry items are still in my inventory. 

 

I have several blogs and a Google Plus page (I am not really a networker -- well I don't think of myself that way -- as I don't Tweet or do Facebook. My Google plus page is mostly about Blender and tutorials and Opensim (where I am no longer so involved). 

My 8 year old fashion blog -- coming up on 4,500 posts is here :    http://chicatphilsplace.blogspot.com/

 

[Drongle McMahon]

"The reason si that it'll take so long to clean up the file after the modifier..."

Are you saying that all Blender modifiers suffer from this "wobble" effect? I haven't investigated that. Auto-smooth, which does, isn't, technically, a modifier. Anyway, I wrote an R function that does the rounding of the normals I described (13,000 or so of them) in about half a second (on ssd disc). That's fast enough for me.

 

daetrim<-function(infile,idkeywd="normal",decdigs=4,outfile=infile){  library(xml2)  adoc<-read_xml(infile)  fanodes<-xml_find_all(adoc,"//d1:float_array")  nodeids<-sapply(xml_attrs(fanodes),function(x){x["id"]})  normidx<-grep(idkeywd,nodeids,ignore.case=TRUE)  for(i in 1:length(normidx)){    norms<-as.numeric(strsplit(xml_text(fanodes)[normidx[i]]," ")[[1]])    norms<-round(norms,decdigs)    anode<-fanodes[normidx[i]]    xml_text(anode)<-paste0(as.character(norms),collapse=" ")  }  write_xml(adoc,outfile)}

In case anyone is unfamiliar with R, you can find it at R-project.org. It's the most commonly used open source software used by academic statisticians, and is especially good at handling large arrays/vectors. I should add some error trapping code to this!

[ChinRey]

---

Drongle McMahon wrote:

Are you saying that all Blender modifiers suffer from this "wobble" effect?

---

I didn't know about the wobble effect until you mentioned it here but Blender has "vertice drift" even for vertices that are supposed to remain unchanged and this drift can often be big enough to affect the LI. The algorithms used by the modifiers don't know about SL's sensitivity to vertice positioning and do not take that into account. The only way to get max. optimisation is to check every vertice manually. On second thought, that might be taking things a bit further than most mesh makers are prepared to go but it's not actually that difficult or time consuming once you've got the routines down and prepare for it all through the modelling process.

 

---

Drongle McMahon wrote:

Anyway, I wrote an R function that does the rounding of the normals I described (13,000 or so of them) in about half a second (on ssd disc). That's fast enough for me.

---

Wow, yes that sounds like a useful tool!

I guess it's time for me to take the plunge and learn the dae syntax. If I only can find the bits about meh models among allt he talk of animations in the docs.

 

[Drongle McMahon]

I'm going to make a better one that can iron out small variations without being as drastic as rounding. Maybe that could do positions as well as normals. It would be better if someone could do it in Python, as anyone with Blender automatically has Python. Even better would be an addon modifier that could be put last on the stack in Blender to do it! Too hard for me at the moment.

[Aquila Kytori]

I was curious about what you meant about not using modifiers because of “vertex drift” so found a little script here :

http://blender.stackexchange.com/questions/1311/how-can-i-get-vertex-positions-from-a-mesh

that prints out vertex  x  y and z positions of meshes in Blender.

I began by: in top view adding a plane and then in edit mode running the script :

Then moving one of the vertices along the x axis by 0.1 :

Accurate to 7 decimal places, I repeated for moving  0.2, 0.3, 0.4, 0.5 etc up to 1.0 along x axis. The 0.5 was exactly 0.5 in the print out but all the rest had a similar deviation as the first. Was this the kind of “drift" that you were referring to ? If so does the collada exporter export to 7 or 8 decimal places ? surely not .

 

Next I cut my plane in half and added a mirror modifier and after applying it ran the script. All the vertices showed no deviation.

Lastly I added a plane in top view, then in:

     Edit mode, rotated it about the x axis 90°

     Front view,   G X 1   followed by   G Z 1, so that one corner of the plane was at x0 y0 z0.

Ran the script and got the following :

 

5.960464477539063e-08  :smileysurprised:   How can that even approximate zero ?  What does this number mean ???

When I Googled for it I found some results that were Blender related but I couldn’t understand them.

Also other none Blender related results ! !  

How can such an apparently crazy number (to me anyways ) get so many results from Google ?

Hoping you , Drongle or some one else with a maths brain can explain it to me.

 

Last thing I did was move that corner vertice along the x axis 0.1m: